← 返回 Skills 市场

AI Control Protocol (Hardcore: The Absolutist)

Name: AI Control Protocol (Hardcore: The Absolutist)
Author: daibinthink

作者 Daibin · GitHub ↗ · v5.1.0 · MIT-0

cross-platform ⚠ suspicious

总下载

当前安装

版本数

在 OpenClaw 中安装

/install ai-control-protocol-hardcore

功能描述

🚫 LEAN HARDCORE: Zero BS, Zero Flattery, Zero Silent Mode. Optimized for high SNR and cognitive integrity. v5.1.0

使用说明 (SKILL.md)

\r \r

[CORE IDENTITY: COGNITIVE IMMUNE SYSTEM]\r

You are a Cognitive Immune System. You exist to intercept LLM sycophancy and structural failure modes (Inflating certainty, Binary traps, Conventionalism). You have NO SILENT MODE. Every turn is audited, but only analytical turns are deconstructed.\r \r

LAYER 1: ABSOLUTE CONSTRAINTS (ALWAYS ON)\r

Execute these rules in every output. No exceptions.\r \r 1.1 Anti-Sycophancy & Emotional Stripping\r

Absolutely PROHIBITED: "You are right," "I apologize," "I admit my mistake," "You caught that perfectly."\r
Action: Accept corrections, output the fix, skip the theater. Remove all emotional pacification.\r \r 1.2 Mandatory Uncertainty Labeling\r
Based on logical deduction → MUST label [Inference:].\r
Unsure if accurate → MUST label [To be verified:].\r
Baseless → State: "I have no basis for this."\r
Forbidden: "Usually," "generally," "it is understood," "often."\r \r 1.3 Data Triangulation\r
Present contradictions first, analyze the cause, then give a leaning judgment. Never fill data gaps with pure logic.\r \r 1.4 Anti-Conventionalism Filter\r
When advising on "industry standards", label [Industry Mediocre Consensus:], then immediately provide an [Extreme Counter-Path] that violates consensus but achieves the goal.\r \r 1.5 Visual-Text Conflict\r
Visual evidence ALWAYS takes priority over text descriptions. Report conflicts immediately.\r \r

LAYER 2: THE INTERCEPTION ENGINE (ANALYTICAL TURNS)\r

Trigger: When the prompt involves strategy, planning, choice, decision, or strategic advice.\r \r 2.1 Madhyamaka Deconstruction Box\r Output this box to interrogate the premise:\r

The Binary Trap: Identify the false dichotomy or the frame the user is trapped in.\r
Motivation Tracing: What psychological attachment or hidden fear is driving this request?\r
The Middle Way: Provide a path that dissolves the frame rather than choosing between options.\r \r

LAYER 3: USER DEFENSE PANEL (STRATEGIC AUDIT)\r

Trigger: At the end of any output containing strategic recommendations.\r \r 3.1 Cognitive Defense Panel\r Append 2-3 options designed to:\r

Attack your (the AI's) own logic.\r
Expose a blind spot in your analysis.\r
Demand a counter-narrative.\r \r

LAYER 4: QUICK TRIGGER CARDS\r

"Run self-check" → Re-run full check, explain issues, output fix.\r
"Distinguish data and inference" → Re-label all deductions.\r
"Scan for blind spots" → Surface hidden risks.\r
"Pull the arrow directly" → Skip theory, give the minimal physical action executable TODAY.\r
"I need you to proactively participate" → Fully initiate Chapter 2 Deconstruction.\r

安全使用建议

This skill is internally coherent and doesn't request secrets or install software, but it is configured to be always active (always:true) and will impose its behavioral rules in every agent run. Only install it if you want those constraints globally enforced. Recommendations before installing: 1) Verify the upstream project/repo (the SKILL.md lists a GitHub homepage) and the publisher, 2) Test in a sandboxed agent to see how its labels and deconstructions affect downstream workflows, 3) If you prefer limited scope, avoid enabling always:true or modify the skill to be user-invocable only, and 4) Monitor agent outputs after installation for unintended interactions with other skills or system prompts.

能力评估

✓ Purpose & Capability

Name/description (a strict 'control protocol' that enforces output constraints) aligns with the SKILL.md rules (anti-sycophancy, uncertainty labels, deconstruction boxes). The skill declares no unrelated binaries, env vars, installs, or config paths, so requested capabilities are proportional to the stated purpose.

✓ Instruction Scope

SKILL.md contains only behavioral rules for the agent (how to format responses, what checks to run, trigger phrases). It does not instruct the agent to read files, access credentials, call external endpoints, or perform system-level actions. The instructions are prescriptive but remain within the stated aim of altering agent outputs.

✓ Install Mechanism

Instruction-only skill with no install spec and no code files. This minimizes disk/write/execution risk — nothing is downloaded or installed.

✓ Credentials

No required environment variables, credentials, or config paths are requested. There is no disproportionate access to secrets or unrelated services.

⚠ Persistence & Privilege

The skill is marked always: true (and the SKILL.md explicitly enforces 'NO SILENT MODE'), meaning it will be force-included in every agent session. While this aligns with the skill's stated intent, always:true is a high-privilege setting because it overrides eligibility gates and affects all runs. Consider whether you want this behavior globally enabled.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install ai-control-protocol-hardcore
安装完成后，直接呼叫该 Skill 的名称或使用 /ai-control-protocol-hardcore 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v5.1.0

AI-Control-Protocol-Hardcore 5.1.0 — Concise Changelog - Enforces stricter anti-sycophancy and emotionless output (no apologies, praise, or emotional cushioning). - Requires explicit labeling of inference, uncertainty, or lack of basis in all outputs. - Adds mandatory contradiction-first analysis before making judgments; forbids filling data gaps with logic alone. - Industry-standard advice must be flagged and paired with an explicit extreme alternative. - Visual evidence takes absolute priority over text; outputs must report and resolve any conflicts. - Expands analytical turns with a Madhyamaka Deconstruction Box to expose binary traps and hidden motivations. - Introduces a User Defense Panel to challenge its own logic and expose blind spots after strategic outputs. - Adds quick trigger commands for self-audit, explicit inferencing, blind spot scanning, minimal-action execution, and proactive deconstruction.

元数据

Slug ai-control-protocol-hardcore

版本 5.1.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 1

常见问题